Corpus 1,272 assessed · 1,173 scored · 643 reproduced ≥75 · 168 flagged ·∅ 74.1/100
← New search

Gene Dosage Analysis on the Single-Cell Transcriptomes Linking Cotranslational Protein Targeting to Metastatic Triple-Negative Breast Cancer.

Pharmaceuticals (Basel) · 2021
L1 53/100 3/4
⚑ Flagged for review — a reproduced result did not match the reported value

Provisional — an automated or curator check raised a specific concern and points reviewers here. This is NOT a final assessment and not a determination about the authors.

Why this verdict

The main results reproduced, with only marginal, non-material deviations.

Reproduced on the brainbox compute brainarbeit.com
Scoring basis — itemised

Every item that counted toward this verdict, and the exact part of the reproduction that produced it.

Supporting (toward a concern)
Content-critical question only partially held
+2 pts
From: Q5 · Derivability / plausibility 🟡
Content-critical question only partially held
+2 pts
From: Q7 · Core claim 🟡
Content-critical question only partially held
+2 pts
From: Q8 · Severity of the miss (overall human judgment) 🟡
Minor / cosmetic deviation
+1 pts
From: Q6 · Severity of the deviation 🟡
Minor / cosmetic deviation
+1 pts
From: Q2 · Endpoint comparability 🟡
Concordant (toward reproduced)
Code + data deposited & functional
-2 pts
From: Data & code availability Available & functional
Total score +6
✓ What held up
  • Nothing in this column.
What did not (or only partly)
  • 🟡Could not use the authors’ exact input data
  • 🟡Reported values were only indirectly comparable
  • 🔴A deviation arose in the data or preprocessing
  • 🔴A deviation was attributed to the published material
  • 🟡Reported values were not (fully) derivable from the shared data
  • 🟡The deviation was non-trivial in magnitude
  • 🟡The central claim did not (fully) hold under reproduction
  • 🟡Overall, the reproduction showed a material discrepancy
How its reproducibility compares
53/100
Reproducibility score
1.2 SD below mean
vs. all fields · 1173 studies
🎯 Scores higher than 14% of all assessed papers rank 1005 of 1173 scored

A 0–100 reproducibility-quality score from the per-question grades, shown as a z-score: standard deviations above (+) or below (−) the mean of comparable assessments.

Reproduction agent’s raw note

Described well enough to VERIFY THE ASSIGNED DATASET, but NOT to reproduce the paper's core pipeline. GSE75688 (the paper's validation set) downloaded + profiled on «infra»: it is a clean, complete, open Smart-seq2 TPM matrix (57,915 genes x 563 expression cols, 11 patients BC01-BC11). Two reported numeric facts about it reproduce EXACTLY by direct inspection/arithmetic: 57,915 genes, and 7,528,950 Z-scores = 57,915 genes x 130 cells (internally consistent; no fabrication evident). The '130 cells'/'5 TNBC patients' are post-filter/external-subtype facts (raw TNBC tumour single cells = 89; subtype labels are not in the deposit). The ONE shipped 'code' artifact is the generic vegan R package, cited only for the Fig 5 diversity indices; running vegan 2.7.1 diversity() on GSE75688 reproduces the qualitative Fig 5B finding (Shannon & Simpson strongly correlate: Pearson 0.86 / Spearman 0.95) -- though Fig 5 in the paper is computed on the PRIMARY dataset, so this is a method cross-check, not a like-for-like number. NOT ATTEMPTED / not reproducible: the central CNV-gene-dosage concordance pipeline and everything derived from it (20,651 CNV events, 86/94 CNG-UP genes, 33 SRP genes, 5 MCODE modules, GO p-values, cBioPortal survival) -- because (a) the authors shipped NO custom code (only a prose Methods + a generic package link), and (b) the CNV modality is entirely absent from GSE75688 and the room was given no matched DNA-seq. Reproducing those would be re-creation from prose, not reproduction, so they are honestly recorded as out-of-scope rather than fabricated. Overall: dataset solid + count claims exact; pipeline non-reproducible from shipped artifacts.

💻 Code ↗ 🗄 Data: GSE75688

These records describe the outcome of reproduction attempts carried out autonomously by brainbox using large language models (LLMs). They are not peer review, not an audit, and not a determination of error or misconduct by any author. A verdict reflects what one attempt could or could not reproduce — which may depend on data access, undocumented parameters, the computing environment, or the depth of effort — and not a judgement of the people who did the work. We can be wrong, and we correct mistakes quickly: every record carries a “report an error” button.

Assessment versions

Every reproduction run is kept as an immutable version — anchored to the data as it stood, with a tamper-evident chain hash. A rerun (e.g. after an author updates a deposit) adds a new version; the previous one stays on record.

  1. v1 current initial assessment Score 53
    assessed: 2026-06-18 ⛓ a8ffe567570a
✎ I am an author of this paper

Updated or fixed a deposit, or is there an erratum? Ask us to re-run the metrics. We verify by email first; the new result is published as a new version with full history — nothing is overwritten.

Reason for the rerun

We email you a confirmation link first. The rerun is an objective re-measurement — it cannot change the verdict in your favour, only ask us to look again.

Provenance — full disclosure

When this reproduction was carried out, which methodology version was used, and by whom — so the record can be audited and checked independently.

Reproduced
2026-06-18
Rubric version
v1.0
Assessed by
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-06-19
no human curator yet
Last updated
2026-08-05

Provisional, curator- or AI-assessed, and independently checkable. A reproduction outcome states what one attempt could reproduce — not a judgement of the authors.

Deep full-text extraction

Model: opus
Founding hypothesis

Can a computational framework integrating concordant copy number variation (CNV) and single-cell gene expression reveal common gene dosage effects across cell types in metastatic triple-negative breast cancer (TNBC), and does amplification-induced upregulation of ribosome/protein-targeting genes drive its metastatic potential?

Core claims
  • A computational framework integrating independently measured CNV (DNA sequencing) and single-cell RNA-seq from the same patients identifies recurrent concordant copy number gain and gene upregulation (CNG-UP) events and functional modules at the single-cell level. method
  • Consistent copy number gains and upregulation in metastatic TNBC are enriched for ribosome proteins involved in SRP-dependent cotranslational protein targeting to membranes. finding
  • SRP-dependent cotranslational protein targeting is the top functional module, validated as the prioritized module in an independent metastatic TNBC dataset. finding
  • Increased ribosome gene copies in TNBC associate with enhanced stemness, differentiation, and EMT/metastatic potential. mechanism
  • The 33 ribosome/protein-targeting genes are frequently mutated in metastatic breast cancers and their mutation associates with worse patient survival. finding
  • Mutations in the 33 genes correlate with adverse clinical features including higher histological grade, later tumor stage, higher aneuploidy and hypoxia scores, and older diagnosis age. finding
  • Ribosome protein modules represent a potential target for TNBC therapy. resource
Experimental setups
Assay System Perturbation Readout Platform
single-cell RNA-seq combined with matched whole-exome/DNA sequencing CNV analysis (computational dosage framework) metastatic TNBC patient tumors, GSE118389/GEO118390 (6 patients, 1533 cells) none (observational CNV) concordant CNG-UP events, Z-scored relative gene expression mapped to CNV
single-cell RNA-seq with matched CNV analysis (independent validation) metastatic TNBC patients, GSE75688 (5 patients, 130 cells) none CNG-UP events and functional modules (SRP-dependent cotranslational protein targeting)
functional enrichment / GO & pathway analysis with MCODE module detection on human interactome 94 top CNG-UP genes from TNBC single cells none enriched GO terms, pathways, and functional modules MCODE algorithm
mutational frequency and survival analysis on bulk cancer genomics data 6688 breast cancer samples across 12-15 studies; survival on 4821 patients none mutation frequency per gene, overall/relapse-free/disease-specific/disease-free/progression-free survival public cancer genomic resource (cBioPortal-type)
intratumor heterogeneity and cell-state analysis (diversity indices + tSNE) 6 TNBC patients, GSE118389 single cells none Shannon-Wiener index, Simpson index, marker-based cell states (stemness, pluripotency, differentiation, proliferation, EMT/metastasis) vs SRP module tSNE
Key results
  • 47,514 CNG-UP events associated with 94 genes across 1145 cells identified in primary TNBC dataset 94 genes; 47,514 events (abstract: 47,198)
  • Cotranslational protein targeting to membranes was the most significantly enriched functional term (20 genes) corrected p = 10×10^-25.133
  • Independent dataset yielded 86 genes with CNG-UP in 6+ cells; confirmed SRP-dependent cotranslational protein targeting as top module 33 overlapping genes (7 in common); 29 ribosome protein genes
  • 31 of 33 genes related to protein translation; 7 related to VEGFA-VEGFR2 signaling p = 10×10^-56.43 (translation); p = 10×10^-3.76 (VEGFA-VEGFR2)
  • Three metastatic breast cancer cohorts highly mutated (>50% patients) in the 33 genes, unlike non-metastatic cohorts >50% of patients
  • Patients with mutations in the 33 genes had shorter median overall survival (145.43 vs 175.30 months) 145.43 vs 175.30 months
  • Ribosome (SRP) module positively associated with cell differentiation, stemness, and EMT/metastasis states
  • Top mutated genes included MRPL13 (17%), SRP9 (15%), PABPC1 (15%), RPL8 (15%) 11-17% mutation frequency
Key statistics
  • count 47,514 CNG-UP events / 94 genes / 1145 cells (primary TNBC dataset filtered concordant events)
  • pvalue corrected p = 10×10^-25.133 (cotranslational protein targeting to membranes enrichment)
  • pvalue 10×10^-56.43 (33 genes protein translation enrichment)
  • pvalue logrank p = 8.94×10^-7, Q = 4.24×10^-6 (overall survival altered vs unaltered group)
  • mean 145.43 vs 175.30 survival months (median OS mutated (1625 patients) vs unmutated)
  • count 6688 breast cancer samples, 12 studies (mutational analysis cohort)
  • count 89,524 meaningful CNV-expression events (|Z|>1.96) of 1,440,802 total in 1245 cells (primary dataset filtering)
  • other mutation frequencies 11-17% (top ten mutated ribosome/SRP genes)

Statistical methods review

Model: sonnet

A neutral, descriptive read of the statistical approach — what was done, and (for shared learning, not as criticism) what could also have been done.

The paper presents a computational framework that maps CNV data from bulk DNA sequencing to Z-score-transformed single-cell RNA expression profiles from the same patients to identify concordant copy number gain and gene upregulation (CNG-UP) events in metastatic TNBC. Primary analysis was conducted on 1145 cells from 6 patients (GSE118389), with independent validation in a second cohort of 5 patients (GSE75688); enriched genes were clustered into functional modules using the MCODE network algorithm. Survival analyses using log-rank tests with Q-value correction were performed on pooled multi-cohort breast cancer data (up to 6688 samples), and intratumor heterogeneity was characterised using Shannon-Wiener and Simpson ecological diversity indices.

Replicationbiological Sample sizePrimary dataset: 6 patients, 1533 cells; validation dataset: 5 patients, 130 cells; survival cohort: pooled 4821–6688 patients across 12–15 studies. No formal power analysis or sample-size justification stated. GroupsCNG-UP concordant events vs. non-concordant expression-CNV pairs across cell types; patients with any mutation in 33-gene module (n=1625) vs. patients with no mutation (unaltered group) for survival Pairingmixed Randomization/blindingnot stated Dispersionnone Exact p-valuesyes Effect sizesno Confidence intervalsno Multiplicity correctionQ-value (FDR-type) correction for survival endpoints; 'corrected p-value' for GO/pathway enrichment — specific algorithm (e.g., Benjamini-Hochberg) not named in either case
Statistical tests used
Test Applied to n Assumptions
Z-score threshold (|Z| > 1.96) applied as a significance criterion for single-cell gene expression relative to the cross-cell mean Identifying significantly up- or downregulated genes in each of the 1533 cells across the primary TNBC dataset; analogous threshold applied in validation dataset 1533 cells from 6 patients (primary); 130 cells from 5 patients (validation) not stated
Gene Ontology and biological pathway functional enrichment analysis with corrected p-values (method not named) Enrichment of 94 CNG-UP genes (primary) and 33 overlapping SRP-module genes across GO terms and pathways including VEGFA-VEGFR2, Rho GTPase, spliceosome, glycolysis 94 genes (primary); 86 genes (validation); 33 genes (overlap) not stated
Log-rank test with Q-value correction Five survival endpoints (overall survival, relapse-free, disease-specific, disease-free, progression-free survival) comparing patients with vs. without mutations in the 33-gene module 4821 breast cancer patients (survival analysis); 6688 samples across 12 studies (mutation frequency) not stated
Shannon-Wiener diversity index and Simpson diversity index Quantifying intratumor expression heterogeneity per patient; correlation between the two indices across patients 6 patients (GSE118389) na
MCODE (Molecular Complex Detection) network clustering algorithm Clustering GO-annotated CNG-UP genes from the human interactome into functional modules (5 modules in primary, 3 in validation) 94 genes (primary); 86 genes (validation) na
tSNE (t-distributed stochastic neighbour embedding) Dimensionality reduction for visualising cell-state variation across all single cells and defining five cell states (stemness, pluripotency, differentiation, proliferation, EMT/metastasis) 1145 cells (primary dataset) na
Approaches that could also have been used
  • Survival associations were assessed with log-rank tests comparing a binary altered/unaltered grouping, and the paper separately documents that grade, stage, age, and chemotherapy receipt differ substantially between these groups
    Could also: Multivariable Cox proportional hazards regression could also have been used, adjusting for the clinical covariates shown to differ between groups — A multivariable Cox model would allow assessment of whether the 33-gene module is independently prognostic beyond stage, grade, and treatment, and would yield hazard ratios with confidence intervals as interpretable, reportable effect sizes
  • Single-cell gene expression significance was determined by a fixed Z-score threshold of |Z| > 1.96, applied uniformly across all genes and cells
    Could also: Statistical frameworks designed for single-cell count data — such as MAST (a hurdle model), DESeq2, or edgeR — could also identify differentially expressed genes while explicitly modelling dropout and the distributional properties of UMI or read counts — Dedicated scRNAseq differential expression tools account for zero-inflation (dropouts) and overdispersion characteristic of single-cell experiments, which the Z-score approach applied here does not model, potentially affecting the set of genes prioritised
  • The recurrence threshold for retaining CNG-UP events was set at ≥100 cells (primary) or ≥6 cells (validation) as absolute counts without a stated statistical rationale for the cut-off values
    Could also: A permutation test or binomial/hypergeometric test could also be used to assess whether the observed co-occurrence of CNG and upregulation in a given number of cells exceeds chance expectation given the marginal frequencies of each event — A significance-based recurrence threshold would provide a statistical basis for the cutoff, yield a p-value or FDR for each candidate gene's recurrence, and allow estimation of false discovery rates among retained CNG-UP events
  • GO and pathway enrichment used a corrected p-value on a threshold-defined binary gene list (94 genes), with the correction method unnamed
    Could also: Gene Set Enrichment Analysis (GSEA) on a continuously ranked gene list (e.g., ranked by recurrence frequency or mean Z-score) could also test for pathway enrichment without requiring a binary threshold — Rank-based enrichment methods use the full score distribution rather than a binary gene list, reducing sensitivity to the choice of cutoff and providing a normalized enrichment score as an effect-size analogue alongside the p-value
  • Clinical associations (race, grade, stage, chemotherapy, diagnosis age, aneuploidy score, hypoxia score) between altered and unaltered patient groups were described narratively with figures but without formal statistical tests or reported p-values
    Could also: Chi-squared or Fisher exact tests (for categorical variables: race, grade, stage, chemotherapy) and Mann-Whitney U tests or t-tests (for continuous variables: diagnosis age, aneuploidy score, hypoxia score) could also formally quantify these associations, with multiplicity correction across the seven features — Formal tests with reported statistics, p-values, and effect sizes (e.g., odds ratios for categorical associations) would allow readers to assess the strength and precision of each observed association and distinguish signal from description
  • Intratumor expression heterogeneity was quantified per patient using Shannon-Wiener and Simpson ecological diversity indices treating expressed genes as analogues of species
    Could also: Purpose-built single-cell heterogeneity metrics such as scEntropy, or variance decomposition via mixed-effects models partitioning expression variance into between-patient and within-patient components, could also quantify transcriptomic diversity — Ecological diversity indices were developed for species-abundance distributions; tools designed for transcriptomic data can account for the specific sparsity, gene-gene correlations, and dropout structure of scRNAseq, potentially yielding more biologically interpretable heterogeneity estimates
Software: MCODE (Molecular Complex Detection) algorithm · Public cancer genomic resource (likely cBioPortal, inferred from mutational and survival data aggregation across 12–15 studies)

What was reproduced

The exact results taken into scope, with each reported value next to the value our attempt produced.

Scope — pmid-34577617

Paper: Liu Y, Zhao M. Gene Dosage Analysis on the Single-Cell Transcriptomes Linking Cotranslational Protein Targeting to Metastatic Triple-Negative Breast Cancer. Pharmaceuticals (Basel) 2021. DOI 10.3390/ph14090918. PMCID PMC8472593.

Assigned data: geo:GSE75688 (Chung et al. 2017 breast-cancer scRNA-seq) — this is the paper's validation dataset. Assigned "code": github.com/vegandevs/vegan — the generic community-ecology R package. The authors shipped NO custom code/scripts. vegan is cited only for the diversity (heterogeneity) indices and an ordination correlation in Fig 5.

Reported results and their pipeline origin

Result (paper) Pipeline In scope? Why
GSE75688 has 57,915 genes; 7,528,950 Z-scores over 130 cells (5 TNBC patients) direct count / Z-score transform YES derivable by direct inspection of the deposited matrix
Z-score transform on primary set: 1,533 cells, 21,785 genes, 33,396,405 Z-scores custom prose pipeline on GSE118389/90 NO (wrong dataset) not GSE75688; primary dataset not assigned to this room
CNV–expression concordance → 89,524 consistent events → 94 CNG-UP genes (primary); 86 CNG-UP genes (GSE75688) custom CNV/Z concordance NO no shipped code; requires matched patient CNV/DNA-seq which GSE75688 (expression-only TPM) does not contain
33 SRP-dependent genes shared; 7 overlapping genes (Fig 2E/2F) MCODE on gene network NO downstream of above; no code; needs the CNG-UP gene lists
5 MCODE functional modules; GO enrichment p-values MCODE / Metascape (external GUI) NO external interactive tool; not a scriptable shipped pipeline
Mutation frequencies (cBioPortal, 6,688 samples); survival logrank p=8.94e-7 cBioPortal queries on external TCGA cohorts NO external cohorts (not GSE75688); GUI/portal, no code
Fig 5A Shannon-Wiener & Fig 5B Simpson~Shannon correlation; Fig 5C t-SNE envfit vegan diversity + ordination PARTIAL the ONE shipped tool; computed on the primary set in the paper (no GSE75688 numbers reported) → we reproduce the method on GSE75688 as a qualitative cross-check

What this room attempts

  1. Dataset profiling of GSE75688 (required) — N genes/cells/patients reported vs observed, QC.
  2. Exact count claims about GSE75688: 57,915 genes; 130 cells (via 57,915×130 = 7,528,950 Z-scores).
  3. Methodological repro of the vegan diversity result (Fig 5A/5B) on GSE75688 tumour single cells: Shannon & Simpson per cell and their correlation — qualitative (paper reports no GSE75688 diversity numbers).

Explicitly NOT attempted (and why)

  • The core CNV-gene-dosage concordance pipeline and all results derived from it (86/94 CNG-UP genes, 33 SRP genes, MCODE modules): no code shipped and the CNV modality is absent from GSE75688. Reproducing it would require re-implementing an under-specified prose pipeline and sourcing matched CNV data the room was not given — that is re-creation, not reproduction.
  • Survival / mutation-frequency analyses: external TCGA/cBioPortal cohorts, GUI-driven.
Figures / tables: Fig 2CFig 5B
C1
Reported
57,915 genes (GSE75688)
Reproduced
57,915 genes (matrix rows)
exact
C3
Reported
7,528,950 Z-scores (GSE75688)
Reproduced
7,528,950 = 57,915 genes x 130 cells
exact
C2
Reported
130 cells (5 TNBC patients)
Reproduced
internally consistent w/ 57,915x130; raw TNBC tumour SC observed = 89
partial
C4
Reported
5 TNBC patients
Reproduced
11 patients total; 5 TNBC (BC07-BC11) per Chung 2017
partial
C7
Reported
Shannon-Wiener & Simpson indices correlate (Fig 5B)
Reproduced
Pearson r=0.861, Spearman rho=0.948 (GSE75688 TNBC tumour SC, vegan 2.7.1)
partial
C5
Reported
20,651 CNV events (GSE75688)
Reproduced
not reproducible (GSE75688 is expression-only, no CNV)
did not match
C6
Reported
86 CNG-UP genes (GSE75688)
Reproduced
not reproducible (no shipped code + no CNV data)
did not match

Assessments & scoring basis

Each contributor’s verdict, the per-question basis, and the auditable, itemised worksheet behind it.

🤖 AI curator · claude (ai-curator room) · v1.0 L1 53/100

An automated assessment. It can flag an open question for review but can never, on its own, record a discrepancy verdict (C5) against a paper.

🟡1. Data identity
🟡2. Endpoint comparability
🔴3. Location of the main deviation
🔴4. Cause of the deviation
🟡5. Derivability / plausibility
🟡6. Severity of the deviation
🟡7. Core claim
🟡8. Severity of the miss (overall human judgment)
Scoring basis — itemised

Every item that counted toward this verdict, and the exact part of the reproduction that produced it.

Supporting (toward a concern)
Content-critical question only partially held
+2 pts
From: Q5 · Derivability / plausibility 🟡
Content-critical question only partially held
+2 pts
From: Q7 · Core claim 🟡
Content-critical question only partially held
+2 pts
From: Q8 · Severity of the miss (overall human judgment) 🟡
Minor / cosmetic deviation
+1 pts
From: Q6 · Severity of the deviation 🟡
Minor / cosmetic deviation
+1 pts
From: Q2 · Endpoint comparability 🟡
Concordant (toward reproduced)
Code + data deposited & functional
-2 pts
From: Data & code availability Available & functional
Total score +6

The clean, checkable facts about GSE75688 reproduce exactly and are internally consistent (57,915 genes; 57,915×130 = 7,528,950 Z-scores), with no fabrication evident. However, the paper's central CNV-gene-dosage pipeline (20,651 CNV events, 86 CNG-UP genes, and all SRP/MCODE/GO/survival downstream) is non-reproducible from the shipped artifacts: no author code was shipped (the code link is the generic vegan package) and the assigned deposit is expression-only with no CNV modality. The deviation therefore sits on the authors'/availability side (incomplete code + a headline number attributed to a dataset lacking that modality), not in a demonstrated computational disagreement. Overall this is partial: solid and exact where checkable, core conclusion untested and explainably non-reproducible — yellow rather than red because nothing is proven false or fabricated.

🤝
Reproduced automatically — and fairly

Automated reproduction checks whether a published result can be regenerated from the paper’s described methods and shared data. When something does not reproduce, that is not a claim of error or misconduct — most often it reflects under-described methods, software or environment differences, or gaps in data access, and some of the pre-print papers in the queue may carry issues their authors had no part in. The goal is shared awareness that rigorous, fully-described methods help everyone — never a judgement of any author.

Are you an author? We would genuinely like to hear from you — to clarify the record, add data or code, re-run the pipeline after an accession update, and publish your response right next to the assessment. Everything here is open and auditable.

🚩 Report an error in this record

Spotted something wrong — a verdict you’d contest, a data or value error, or a private detail that slipped through? Tell us, with a short justification. Authors and readers are equally welcome to write in; we review every report.

Prefer email, or the form below not working? Contact us at support@doesitreproduce.com.

Reproduction footprint

claude-opus-4-8

Measured resources invested to assess this paper — sanitised (machine class only, no job ids/paths). Compute = HPC accounting (SLURM); tokens = the AI agent's session.

124.6 k
tokens (I/O) · 5.7 M incl. cache
23 min
runtime · 0 CPU-h
1.7 GB
peak RAM
1
HPC jobs
hummel
machine